August 11, 2026

Nvidia Just Turned GPUs Into Collateral

The question that has haunted every major lender eyeing the AI boom is simple: what happens when the collateral becomes obsolete before the loan matures? On Monday, Nvidia decided to answer it directly, and the answer cost the stock more than 2%.

The Big Question

Nvidia announced strategic partnerships to establish independent compute financing platforms with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR to mobilize over $500 billion of third-party capital for the buildout of AI infrastructure over time. The ambition is real. The structural tension underneath it is larger than the headline suggests.

Why Wall Street Cares

The effort aims to mobilize more than $500 billion in third-party capital for hyperscalers, frontier AI labs and enterprises to build out data centers and acquire Nvidia hardware, marking a potentially important shift in how AI infrastructure is funded. By using institutional credit, insurance funds and private capital to underwrite GPUs and data centers, Nvidia is helping its end users secure financing without tapping their own balance sheets.

The move comes at a specific moment in the AI buildout cycle. In 2025, datacenter capacity was the bottleneck for AI compute growth. By early 2026, the datacenter supply situation improved considerably, but chip production became the limiting constraint. Now, mid-year, it is clear that financing will be one of the most significant obstacles to ramping large-scale compute broadly available to everyone.

Huang said during a February 2026 CNBC interview that AI infrastructure spending could reach $3 trillion to $4 trillion annually by the end of the decade. At that scale, no single company’s balance sheet works. Not even Nvidia’s.

The Bull Case

Jensen Huang’s core argument is a classification upgrade. In a February 2026 CNBC interview, Huang argued that GPUs and AI systems are increasingly revenue-generating assets, and that compute is starting to be treated more like infrastructure.

The market evidence he cites for durability is rental pricing. H100 1-year contract pricing rose from about $1.70 per GPU-hour in October 2025 to about $2.35 by March 2026. But the specific claim that “B200 capacity now runs between $5.30 and $7.05 per GPU-hour” is not something I can verify reliably, so it should be treated as illustrative rather than definitive.

BlackRock CEO Larry Fink has argued that computing power is becoming a distinct asset class, and in May 2026 he spoke about a potential futures market for compute. Apollo has also emphasized the physical scarcity dimension of the compute supply chain.

To make the underwriting viable, Nvidia is putting skin in the game. Some reporting around the announcement described Nvidia as being in talks to guarantee financing in ways that would reduce lenders’ residual value risk, but the specific claim that “Nvidia will guarantee up to 25 percent of the residual value of its chips in individual financing deals” is not confirmed in Nvidia’s own announcement and should be framed as discussion, not as a settled term.

The Bear Case

The skeptics have a sharper argument. The fundamental problem with GPU-backed lending is that semiconductor hardware depreciates on a curve dictated by innovation cycles, not by physical wear. An H100 GPU does not wear out in the traditional sense. It becomes obsolete. And the company best positioned to accelerate that obsolescence is the same company backstopping the loans.

The very company whose chips serve as collateral for these loans is the one releasing the next-generation product that destroys the collateral’s value. That structural conflict is not hypothetical. But the specific claim that “the introduction of Nvidia’s Blackwell chip cratered the resale value of every H100 in the field” overstates what can be supported. Secondary markets have shown volatility and price resets as new generations approach, but the magnitude and universality of that impact varies by form factor, warranty, buyer restrictions, and availability.

Investor Michael Burry has put a number on the risk. In a social media post reported in late 2025, Burry called hyperscaler depreciation practices “one of the more common frauds of the modern era,” arguing that Nvidia’s two-to-three-year upgrade cycle makes five-to-seven-year useful-life assumptions unrealistic, and estimated depreciation could be understated by roughly $176 billion between 2026 and 2028 alone.

The debt supply picture adds another layer. In June 2026, Reuters reported that Morgan Stanley projected worldwide AI-linked debt issuance could reach nearly $570 billion in 2026, compared to roughly $236 billion as of May 31. An increase in supply may lead to wider spreads, even if demand for projects remains high.

The Evidence

The financing facilities will be a boon for startups and mature companies alike, but will also likely fuel concerns from some industry participants that Nvidia is fencing its customers into its own ecosystem. The financing packages will keep a huge amount of capital tethered to Nvidia, and away from fledgling or established competitors.

The precedent for smaller operators already exists. SemiAnalysis has described Nvidia’s backstop program as typically six years in length, during which Nvidia stands ready to purchase compute at pre-agreed price levels that vary over the time period. In that framing, Nvidia provides a take-or-pay commitment to neoclouds, a minimum revenue guarantee on the underlying GPU capacity, and shares in a portion of the neocloud’s revenue earned above the backstop level. The new consortium formalizes and scales this mechanism with institutional balance sheets behind it.

The lenders have structured around the depreciation risk in at least some cases. Apollo has emphasized the physical scarcity of the compute supply chain and the importance of chips, memory, and power as constraints, but specific deal terms like “SOFR plus 400 basis points” and “minimum cash reserves of $100 million” are not consistently documented in public primary materials, so they should not be stated as universal covenant standards.

The Mavens’ View

The most serious institutional investors are not treating this as a simple AI-demand bet. They are asking a more precise question: who absorbs the depreciation risk, and at what price? The MOUs are still subject to definitive agreements. That detail matters more than the $500 billion headline.

Nvidia’s buyback and offtake-style guarantees exist to solve a specific technical problem, GPU residual value, that many lenders are not equipped to price. The bulls lean on this hard. The bears point out that being the rational backstop and being right about residual values are two different things.

What the Morgan Stanley and Apollo projections agree on is scale. Morgan Stanley has framed AI financing needs as enormous, and Apollo’s thematic research has emphasized that physical constraints across the supply chain are likely to persist. The $500 billion figure is not a ceiling. It is an opening position.

What Investors Are Missing

The ecosystem lock-in dimension of this deal receives far less attention than the financing headline. The initiative aims to turn Nvidia compute and full-stack AI infrastructure into an investable asset class, broadening access to AI factories and tying returns to a long cycle of physical construction and power demand. Capital directed through Nvidia-designed financing platforms flows overwhelmingly into Nvidia-designed infrastructure. AMD, Broadcom, and every custom silicon challenger face a competitor that now controls not just the hardware market but the credit conditions for buying into it.

Nvidia describes its compute platforms as investable assets tied to token costs, revenue potential, service life and an ecosystem built on the CUDA platform. CUDA is doing double duty: it is the software lock-in that built the moat, and now it is also part of the underwriting rationale for the debt. That combination has little precedent in technology finance.

The second thing most investors are not fully pricing: the GPU debt cliff is the risk that AI cloud providers and infrastructure partners finance large GPU fleets with debt, then face refinancing, depreciation or utilization pressure as newer Nvidia platforms arrive. If the Rubin architecture, Nvidia’s next-generation roadmap, delivers a step-change in efficiency on schedule, the economic life of Blackwell-era clusters compresses. That is when any residual value guarantees, whatever their final form, get tested.

Stocks to Watch

Nvidia (NVDA): The stock fell on announcement day, which is telling. The market read the consortium as confirmation that Nvidia has reached the outer limit of what it can finance alone. That is true, and it does not change the demand story. Nvidia’s next earnings are scheduled for August 26, 2026.

Apollo Global Management (APO): Shares climbed on the news, the right reaction. Apollo is structured for exactly this kind of long-duration, asset-backed lending and was already among the most active financiers of AI infrastructure before the Nvidia announcement. The consortium gives it a preferred-access channel into the largest single demand pool in private credit.

Blackstone (BX): Up on the day, and with reason. But the specific claim that “Apollo and Blackstone have already structured debt and equity financing for companies including Anthropic” is not something I can verify cleanly in public, so it should be treated cautiously unless supported with a named deal and date. The broader point still stands: Blackstone has deep exposure to data center real estate and power-adjacent infrastructure, making the vertical integration logic unusually clean.

AMD (AMD): The financing consortium is the least-discussed competitive threat to AMD’s momentum. The financing packages will keep a huge amount of capital tethered to Nvidia, and away from fledgling or established competitors. But the specific claim that “GPU revenue [is] forecast to grow 114% in 2026” is not verified here and should be removed or sourced before publication. Preferential financing terms tied to Nvidia hardware still raise the cost of switching for any customer considering alternatives. The moat just got wider.

CoreWeave (CRWV): The neocloud sector is the most direct beneficiary of expanded financing access. CoreWeave built its business model around exactly the kind of GPU-fleet financing this consortium formalizes. Wider availability of institutional capital at better rates reduces the structural disadvantage smaller cloud operators face versus hyperscalers. Watch how the deal terms translate into CoreWeave’s next debt raise.

Nvidia Just Turned GPUs Into Collateral